dl-docker

Docker image for Deep Learning

Pre-requisites

• Docker-CE: Install docker-ce by following the instructions here: https://docs.docker.com/install/linux/docker-ce/ubuntu/
• NVIDIA runtime: Run the setup file (Ubuntu/Centos7) from this repository.

Installed libraries

• CUDA Toolkit 9.2
• CUDNN Library 7.3.1
• Tensorflow 1.12
• Keras latest

Installation

Using Dockerfile

• Clone this repository

git clone https://github.com/hassanmohsin/dl-docker.git
cd dl-docker

• Run the Dockerfile to create an image

docker build -t hassanmohsin/dl-docker:gpu .

Using Docker Hub

• Pull the docker image from the Docker HUB

docker pull hassanmohisn/dl-docker:gpu

Running docker image

• Test the image by running the test script (benchmark.py)

docker run -it --rm -v `pwd`:`pwd` -w `pwd` --runtime=nvidia hassanmohsin/dl:gpu python benchmark.py gpu 20000

• -i (interactive) flag to keep stdin open and -t to allocate a terminal
• --rm to remove the container after executing the script
• -v `pwd`:`pwd` to mount the current working directory to the container with the same path
• -w `pwd to get the working the directory inside the container
• --runtime=nvidia to activate the nvidia runtime

Running Jupyter Notebook

docker run -it --rm --runtime=nvidia -p 8888:8888 hassanmohsin/dl-docker:gpu

This command is to listen to the port 8888 of the docker and forwarding that port through SSH.